aboutsummaryrefslogtreecommitdiff
path: root/embed_internal.h
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-14 15:19:12 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-14 15:19:12 +0200
commitf463018cd700690336898d109226d489deca036d (patch)
tree4e1d83f55a5e231af519aa4d647141db5de23af7 /embed_internal.h
parentd98ee1568c0bd88304a12e59d12104066e01e8d0 (diff)
downloadffwd-f463018cd700690336898d109226d489deca036d.tar.gz
ffwd-f463018cd700690336898d109226d489deca036d.zip
all: add support for Qwen3-Embedding + causal GQA attention
Diffstat (limited to 'embed_internal.h')
-rw-r--r--embed_internal.h6
1 files changed, 6 insertions, 0 deletions
diff --git a/embed_internal.h b/embed_internal.h
index a7365a2..60fc581 100644
--- a/embed_internal.h
+++ b/embed_internal.h
@@ -29,12 +29,18 @@ typedef struct {
const float *norm; /* [hidden] */
} embed_weights_t;
+typedef void (*embed_attention_fn)(
+ float *out, const float *Q, const float *K, const float *V,
+ const int *offsets, int batch, int n_heads, int n_kv_heads,
+ int head_dim, float scale, float *scratch, size_t scratch_bytes);
+
struct embed_model {
embed_config_t config;
embed_weights_t weights;
void *safetensors; /* multi_safetensors_t*, keeps mmap alive */
int layer_start; /* loaded transformer range, inclusive */
int layer_end; /* loaded transformer range, exclusive */
+ embed_attention_fn attention;
};
struct embed_workspace {